TensorFlowLite.md

Integration with TensorFlow Lite on Android

Architecture

Ethan and the car               SSD-MobileNetv2/TensorFlowLite
are on Unity                    runs on Android for object detection
 [Unity] <----> [mosquitto] <----> [Android]
                MQTT server
                username: simulator
                password: simulator

How it works

There are two cameras running on Unity to capture image:

• RGB camera
• Depth camera

These two cameras use "Render Texture" to capture images.

JPEG image data is transferred to an app on Android via MQTT server.

The app on Android receives JPEG data, converts it into Bitmap, then perform object detection by using a pre-trained SSD-MobileNetv2 model on TensorFlow Lite.

The app on Android also supports remote control for the cameras on Unity: pan, tilt and zoom (PTZ).

Code

• => Code on Unity
• => Code on Android